Alexander Suslov commited on
Commit
fe0f47c
1 Parent(s): 71156c8

added mvtec_capsule.py

Browse files
Files changed (1) hide show
  1. mvtec_capsule.py +127 -0
mvtec_capsule.py ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ from enum import Enum
3
+ from pathlib import Path
4
+
5
+ import datasets
6
+ from pandas import DataFrame
7
+
8
+ _HOMEPAGE = "https://www.mvtec.com/company/research/datasets/mvtec-ad"
9
+ _LICENSE = "cc-by-nc-sa-4.0"
10
+ _CITATION = """\
11
+ @misc{
12
+ the-mvtec-anomaly-detection-dataset,
13
+ title = { The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection },
14
+ type = { Open Source Dataset },
15
+ author = { Paul Bergmann, Kilian Batzner, Michael Fauser, David Sattlegger, Carsten Steger },
16
+ howpublished = { \\url{ https://link.springer.com/article/10.1007%2Fs11263-020-01400-4 } },
17
+ url = { https://link.springer.com/article/10.1007%2Fs11263-020-01400-4 },
18
+ }
19
+ """
20
+
21
+
22
+ class LabelName(int, Enum):
23
+ NORMAL = 0
24
+ ABNORMAL = 1
25
+
26
+
27
+ class MVTECCapsule(datasets.GeneratorBasedBuilder):
28
+ """satellite-building-segmentation instance segmentation dataset"""
29
+
30
+ VERSION = datasets.Version("1.0.0")
31
+ _URL = "https://huggingface.co/datasets/alexsu52/mvtec_capsule/resolve/main/capsule.tar.xz"
32
+
33
+ def _info(self):
34
+ features = datasets.Features(
35
+ {
36
+ "image": datasets.Image(),
37
+ "mask": datasets.Image(),
38
+ "label": datasets.ClassLabel(names=["normal", "abnormal"]),
39
+ }
40
+ )
41
+ return datasets.DatasetInfo(
42
+ features=features,
43
+ homepage=_HOMEPAGE,
44
+ citation=_CITATION,
45
+ license=_LICENSE,
46
+ )
47
+
48
+ def _split_generators(self, dl_manager):
49
+ folder_dir = dl_manager.download_and_extract(self._URL)
50
+ category_dir = os.path.join(folder_dir, "capsule")
51
+ return [
52
+ datasets.SplitGenerator(
53
+ name=datasets.Split.TRAIN,
54
+ gen_kwargs={
55
+ "category_dir": category_dir,
56
+ "split": "train",
57
+ },
58
+ ),
59
+ datasets.SplitGenerator(
60
+ name=datasets.Split.TEST,
61
+ gen_kwargs={
62
+ "category_dir": category_dir,
63
+ "split": "test",
64
+ },
65
+ ),
66
+ ]
67
+
68
+ def _generate_examples(self, category_dir, split):
69
+ extensions = (".png", ".PNG")
70
+
71
+ root = Path(category_dir)
72
+ samples_list = [(str(root),) + f.parts[-3:] for f in root.glob(r"**/*") if f.suffix in extensions]
73
+ if not samples_list:
74
+ raise RuntimeError(f"Found 0 images in {root}")
75
+
76
+ samples = DataFrame(samples_list, columns=["path", "split", "label", "image_path"])
77
+
78
+ # Modify image_path column by converting to absolute path
79
+ samples["image_path"] = samples.path + "/" + samples.split + "/" + samples.label + "/" + samples.image_path
80
+
81
+ # Create label index for normal (0) and anomalous (1) images.
82
+ samples.loc[(samples.label == "good"), "label_index"] = LabelName.NORMAL
83
+ samples.loc[(samples.label != "good"), "label_index"] = LabelName.ABNORMAL
84
+ samples.label_index = samples.label_index.astype(int)
85
+
86
+ # separate masks from samples
87
+ mask_samples = samples.loc[samples.split == "ground_truth"].sort_values(by="image_path", ignore_index=True)
88
+ samples = samples[samples.split != "ground_truth"].sort_values(by="image_path", ignore_index=True)
89
+
90
+ # assign mask paths to anomalous test images
91
+ samples["mask_path"] = ""
92
+ samples.loc[
93
+ (samples.split == "test") & (samples.label_index == LabelName.ABNORMAL), "mask_path"
94
+ ] = mask_samples.image_path.values
95
+
96
+ # assert that the right mask files are associated with the right test images
97
+ if len(samples.loc[samples.label_index == LabelName.ABNORMAL]):
98
+ assert (
99
+ samples.loc[samples.label_index == LabelName.ABNORMAL]
100
+ .apply(lambda x: Path(x.image_path).stem in Path(x.mask_path).stem, axis=1)
101
+ .all()
102
+ ), "Mismatch between anomalous images and ground truth masks. Make sure the mask files in 'ground_truth' \
103
+ folder follow the same naming convention as the anomalous images in the dataset (e.g. image: \
104
+ '000.png', mask: '000.png' or '000_mask.png')."
105
+
106
+ if split:
107
+ samples = samples[samples.split == split].reset_index(drop=True)
108
+
109
+ for idx in range(len(samples)):
110
+ image_path = samples.iloc[idx].image_path
111
+ mask_path = samples.iloc[idx].mask_path
112
+ label_index = samples.iloc[idx].label_index
113
+
114
+ with open(image_path, "rb") as f:
115
+ image_bytes = f.read()
116
+
117
+ if mask_path:
118
+ with open(mask_path, "rb") as f:
119
+ mask_bytes = f.read()
120
+ else:
121
+ mask_bytes = bytes(len(image_bytes))
122
+
123
+ yield idx, {
124
+ "image": {"path": image_path, "bytes": image_bytes},
125
+ "mask": {"path": mask_path, "bytes": mask_bytes},
126
+ "label": label_index,
127
+ }